基于Kalman滤波和直方图匹配的双目视觉跟踪  

Binocular Visual Tracking Based on Kalman Filter and Histogram Matching

作　　者：王江[1] 付强[1] 全权[1,2] 蔡开元[1] 

机构地区：[1]北京航空航天大学自控控制系 [2]北京航空航天大学虚拟现实技术与系统国家重点实验室

出　　处：《新型工业化》2013年第2期23-33,共11页The Journal of New Industrialization

基　　金：教育部博士点基金项目资助(20111102120008)

摘　　要：视觉跟踪是计算机视觉领域一个十分重要的问题,目前大部分的研究关注于单目摄像机目标跟踪。而基于单目摄像机的目标跟踪会丢失深度信息,使得当目标出现遮挡、阴影等问题时跟踪难度进一步加大。因此本文提出了一种基于Kalman滤波和直方图匹配的双目视觉跟踪方法。该方法首先利用Kalman滤波预测出目标所在位置,其次利用直方图匹配在局部区域搜索到目标,最后利用极线约束来对匹配结果进行确认,实现完整的预测-匹配-确认的跟踪过程。实验表明,对比Mean Shift算法,该算法能有效减少跟偏或者跟丢的情况,取得了良好的跟踪效果。Visual tracking is a very important problem in the computer vision. Most of the studies focus on the monocular camera target tracking. the object tracking methods based on monocular camera would lose the depth information, which will cause much more difficult to track the object when faced with occlusion, shadow and so on. This paper proposes an algorithm of binocular vision tracking based on Kalman filter and histogram matching. This algorithm firstly uses Kalman filter to predict the position of the target. Then histogram matching is used to search the object in the local region. At last epipolar constraint will be adapted to verify the tracking result at last, which realizes a complete prediction-matching-verification tracking process. The experiments show the algorithm could effectively reduce tracking failure and get better tracking performance compared with the Mean Shift algorithm.

关 键 词：KALMAN滤波 直方图匹配 双目视觉跟踪 极线约束 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术] TH7[机械工程—仪器科学与技术]